FROM BREAST PUMP TO BEST PUMP: A HUMAN-CENTERED EVALUATION

Mothers can experience breastfeeding challenges, and the breast pump is often at the center. Existing literature outlines the range of mothers’ negative experiences with breast pumps, though there is a gap in which breast pump characteristics are important to mothers. Identifying which breast pump characteristics (i.e., portability, ease of use, low-weight, fast milk extraction, comfortability, low-noise, discreet) are important to breast pumping mothers, and whether or not this importance varies between mothers who do or do not work outside of the home will help identify user needs. Collecting user needs informs future breast pump designs in a user-centered design process. A survey collected information on mothers’ experiences with breast pumps and which breast pump characteristics mothers considered important. Summary statistics were analyzed for mothers who did and did not work outside the home, and Latent Class Analysis (LCA) was used to determine whether there were possible groupings between the importance of these characteristics. Summary statistics indicated that mothers considered all seven breast pump characteristics important except for discreet. The only characteristic found as statistically significantly different between mothers of different work statuses was portability. LCA identified a twoclass model with mothers’ age as a significant covariate. Mothers’ work status was not a significant covariate but did predict class membership when considered as a grouping variable in conjunction with mothers’ age. Breast pumping mothers’ needs differ beyond their work status, and collecting and considering these different needs is vital to creating redesigns that improve mothers’ breast pumping experience.

: Class membership probabilities (γ-estimates) and item response probabilities (ρ-estimates) for 2-class model for response Important………17      The manuscript has been prepared for submission to the Journal of Human Lactation using APA formatting with a word limit of 3500 words. This paper will be submitted to the journal by Fall 2019.

BACKGROUND
Breastfeeding is widely recognized as the preferred way to feed and nourish infants. The American Academy of Pediatrics (AAP) recommends breastfeeding for at least the first year of an infant's life, and the World Health Organization (WHO) encourages continuation until two years or longer with complementary foods (World Health Organization, 2019; "Breastfeeding and the Use of Human Milk", 2012). The long-term benefits of breastfeeding to the infant (e.g., stronger immune systems, fewer ear infections, lower rates of heart disease and diabetes) are well established (DiTomasso & Paiva, 2017). Breastfeeding benefits also extend to the mother; studies show a reduced risk of ovarian and breast cancer (Hildebrand, Gapstur, Campbell, Gaudet & Patel, 2013;Su, Pasalich, Lee & Binns, 2013). Importantly, a negative breastfeeding experience is predictive of depressive symptoms in the postpartum phase (Brown, Rance & Bennett, 2015). Thus, while there are comprehensive benefits to a positive, productive breastfeeding relationship between mother and infant, there are often challenges to building and maintaining this relationship. At the nexus of these challenges often lies the human-machine interaction between lactating mother and the breast pump.
A breast pump is a Class I (manual) or Class II (electric) medical device that allows lactating mothers around the world to express and collect their breast milk for future use (Eglash & Malloy, 2015). Surveys show that the majority of breastfeeding mothers prefer electric breast pumps over manual breast pumps or hand milk expression (Clemons & Amir, 2010). Studies show that most breastfeeding mothers in the United States (U.S.) now feed their infants bottled human milk expressed from breast pumps (Felice & Rasmussen, 2015;Labiner-Wolfe, Fein, Shealy & Wang, 2008). Specifically, a longitudinal U.S. survey that followed about 2,000 motherinfant pairs from 2005 to 2007 revealed that 92% of breastfeeding mothers pumped milk at some point in the first year postpartum (Fein et al., 2008). Breast pumping allows breastfeeding mothers to stimulate, extend and/or maintain their capability and effectiveness of extracting their milk faster than manual expression (Eglash & Malloy, 2015;Rasmussen & Geraghty, 2011). Ability to extend or maintain expressing milk quickly is important for women who return to work outside of the home, deal with complications of breastfeeding (e.g., oral thrush, engorgement), care for premature or ill infants, and encourage partner or caregiver participation in feeding (Eglash & Malloy, 2015). Research findings have indicated, however, that significant problems exist with breast pumps, particularly when it comes to comfort, experience, and usability of these devices, which may cause women to end breastfeeding earlier than they had otherwise planned (World Health Organization, 2019;Brown et al., 2015;Dietrich Leurer & Misskey, 2015;Hurst, Engebretson & Mahoney, 2013;Labiner-Wolfe et al., 2008).
Interviews with new mothers showed mothers' attitudes and perceptions towards pumping and its related tasks (e.g., sanitization) to be widely negative (Avishai, 2004;Felice et al., 2017;Hurst et al., 2013). Many women resented the time spent at the breast pump, while in direct contrast, mothers considered feeding at the breast well-spent bonding time with their infant (Avishai, 2007;Felice et al., 2017). In a qualitative analysis of the breast pump experiences of over 1100 women, hundreds of women reported feeling distress, anxiety, pain, and isolation while using a breast pump, which supports the literature (Clemons & Amir, 2010;D'Ignazio, Hope, Michelson, Churchill & Zuckerman, 2016;Flaherman, Hicks, Huynh, Cabana & Lee, 2014;Qi, Zhang, Fein, Wang & Loyo-Berrios, 2014;Tucker, Wilson & Samandari, 2011). The top negative words associated with the use of a breast pump included, "hate," "pain," and "difficult" (D'Ignazio et al., 2016). In addition to emotional and physical distress, the literature describes that mothers find pumping milk to be both labor-intensive and time-consuming (Avishai, 2004(Avishai, , 2007D'Ignazio et al., 2016;Felice et al., 2017). Lack of usability and resultant excessive time commitments are exacerbated when breast pumping mothers work outside of the home. To quote one breast pumping mother who works outside of the home, "the setup is a hassle. Getting the tubes set up, getting everything together, doing it, putting it back, washing it. From start to finish, it takes about 20 minutes… I'm right in the middle of something. Or I can't schedule meetings." (Avishai, 2004). Additionally, mothers are often embarrassed by the distinct look and noise of their breast pumps in the workplace environment (Avishai, 2004;Spitzmueller et al., 2015).
With literature reporting widespread negative experiences with breast pumps, it is essential to understand and consider mothers' comfort and usability of current breast pump designs. This will inform breast pump design changes that will substantially improve the mother-infant breastfeeding relationship, leading to lasting benefits for both parties. A proven method to improve usability and user experience in product design is the user-centered design process. The user-centered design (UCD) process is an evidence-based, iterative approach that considers the end-user's needs, perspectives, and experience to inform the design of a product or system (D'Ignazio et al., 2016;McCurdie et al., 2012;Norman, 2013;"User-Centered Design Basics | Usability.gov", 2019). User-or human-centered design inverts the traditional humanmachine relationship by suggesting that technologies must adapt to match humans instead of humans adapting to technologies (D'Ignazio et al., 2016). More specifically, UCD is a cyclical approach that seeks to identify and understand users and their needs, and meet these needs through design iterations ("User-Centered Design Basics | Usability.gov", 2019). The UCD process has proven beneficial across multiple domains as it identifies challenges early in the design process allowing for quicker solutions, avoids poorly defined system requirements, improves performance by reducing number of user errors, and results in products that actually meet user's needs ("Benefits of User-Centered Design | Usability.gov.", 2019). The literature outlining mothers' negative experiences with breast pumps accentuates the opportunity for applying a user-centered design process to redesign breast pumps that improve usability and the mother's comfort.
The female body has long been cross-culturally considered taboo, which experts argue has directly limited the development of women's health (Almeida, Comber & Balaam, 2016;Rossmann, 2008). In the last eight years, there has been a movement in the field of user-centered design to be more inclusive of women's issues, known as "feminist design" (Bardzell & Bardzell, 2010). This means not only promoting women's active participation in the design process through designing, providing (often overlooked) user perspectives, and beyond, but earnestly investigating the domain of women's health issues in order to understand and design for opportunities in this space (Buckley, 1986;Rossmann, 2008). The breast pump is a prime product for redesign via the application of a feminist design philosophy. With little to no consideration for the experiences of the women who use breast pumps, the design has evolved little beyond a technology that "gets the job done" (D'Ignazio et al., 2016). Incorporating women's experiences and keeping women's health at the forefront of design are necessities to providing adequate, equitable care globally. In this way, feminist design is a vehicle to bring women's health, intimate care, experiences, and needs into prominence.
The literature documents that breast pumping mothers experience both emotional and physical issues with breast pumps which can negatively impact the mother-infant breastfeeding relationship. However, there is an established gap in the literature around what pumping mothers want, need, and desire when it comes to breast pumps. In a user-centered product design process, this is a critical first stepevaluating the needs of the target user groups. In a feminist product design process, determining what is important to mothers about breast pumps informs where the focus should be when redesigning breast pumps that empower mothers and respect their experience. This paper aims to fill this gap in the literature by asking mothers directly what characteristics are important to them in a breast pump. This user-centered design approach will help bring breast pumps out from the shadows of the past and explode into an equitable future. The research presented here seeks to answer two specific research questions: (1) How important are the breast pump characteristics of portability, ease of use, low-weight, fast milk extraction, comfortability, low-noise and discreet to breast pumping mothers? and (2) Are there significant differences in the importance of these breast pump characteristics to breast pumping mothers who work outside of the home versus breast pumping mothers who do not work outside of the home?

Data Collection
A 19-item questionnaire surveyed lactating mothers to gain insights into their experiences associated with breast pumps. To formulate critical questions, market research gathered information on commercially available breast pumps and a literature review established an initial understanding of the issues women may experience with breast pumps. The survey's design and questions went through multiple iterations that were corroborated by industry experts, such as the South County Hospital lactation consultants. Once the IRB was approved (HU1617-125), the consent form and survey questions were facilitated on SurveyMonkey® (see Appendix A). In order to reach a difficult population of current lactating mothers, the link was posted publicly on Facebook, specifically on the South County Hospital New Mothers' Support Group page.

Sample
The survey received 352 responses between March and July in 2017. Eightyseven respondents indicated they had not used a breast pump (at all, or for their most recent baby), so they were omitted from the analysis. One additional respondent indicated she was uncomfortable speaking English; thus, the response to the in-English survey was excluded. These removals resulted in an analytic dataset of n=264 survey respondents. Table 1 outlines the sample characteristics.

Dependent variables
The dependent variables were the self-selected importance levels of seven breast pump characteristics (i.e., portability, ease of use, low weight, fast milk extraction, comfortability, low noise, discreet) to breastfeeding mothers. Survey question 7 asked mothers to independently rate the importance of each characteristic using a five-point Likert scale that ranged from (1) "Unimportant," (2) "Somewhat unimportant," (3) "Somewhat important," (4) "Important," and (5) "Very important."

Independent variables
The independent variable was whether respondents worked outside of the home (survey question 4). After reviewing respondents' comments, three responses were adjusted based on misinterpretations of the question. These three edits resulted in 209 (79%) respondents who work outside the home (WOH), and 55 (21%) respondents who do not work outside the home (NWOH).

Covariates
Demographic characteristics of mothers' work status, age, education level, and manufacturer and model of their breast pump were evaluated as covariates in the statistical analyses. Additionally, demographic characteristics were used as grouping variables in the latent class analysis.

Statistical Analysis
Before beginning statistical analyses, the seven breast pump characteristics were tested for collinearity, meaning one characteristic would predict or explain another requiring omission. No characteristics were found to be collinear (Appendix B); thus, all characteristics were included in further analyses.
Summary statistics were used to determine, overall, which breast pump characteristics were important to mothers. The five-point Likert scale of importance in Question 7 was dichotomized into "Little to no importance" (Likert levels 1, 2 and 3) or "Important" (Likert levels 4 and 5). Percentages then determined the importance of characteristics to the analytic dataset of breast pumping mothers (n=264), WOH mothers (n=209), and NWOH mothers (n=55). Kruskal-Wallis pairwise comparison tests determined whether there were statistically significant differences between the importance of these seven breast pump characteristics between WOH and NWOH mothers (Kruskal & Wallis, 1952).
To further explore whether differences in the importance of breast pump characteristics exist between WOH and NWOH mothers latent class analysis (LCA) was employed. LCA is a statistical method of identifying hidden groups of individuals based on their responses to a set of observed categorical variables. LCA estimates two functional parameters: γ-parameters, probabilities of membership to a specific class, and ρ-parameters, item-response probabilities conditional on class membership (Lanza, Collins, Lemmon, & Schafer, 2007;Lanza, Tan, & Bray, 2013;Miaskowski et al., 2015). A sequence of models was fit with increasing numbers of classes, and various model selection tools were considered in conjunction with the selection of the optimal model. These tools included the likelihood-ratio G 2 statistic (compares expected to observed response pattern proportions), Akaike's Information Criterion (AIC), and Bayesian Information Criterion (BIC) (Akaike, 1974;Schwarz, 1978 instead of a "local" maximum (Dziak & Lanza, 2015). Model interpretability means that each class should be distinguishable from all others, no class should be trivial in size, and it should be possible to assign a meaningful title to each class (Lanza et al., 2007).
After selecting the best-fitting, appropriate model, the model was expanded to include covariates and grouping variables independently. LCA with covariates extends the model to include predictors of class membership, and LCA with grouping variables is a model in which the γ-parameters and ρ-parameters are influenced by membership in an observed group (Lanza et al., 2007). Initial analyses were conducted using R 3.5.1. LCA analyses were conducted using the PROC LCA command in SAS 9.4 (see Appendix C for R and SAS code).

LCA Baseline Model
A series of models with one to five latent classes were fit based on responses of importance. Additional classes were not considered as classes became trivial in size.
The γ-parameters and ρ-parameters were considered per model. Starting with the twoclass model, Table 3 shows the γ-parameter and ρ-parameter estimates. Interpreting Table 3 indicates that 45% of respondents are expected to belong to Latent Class 1, with practically no probability (0%) of considering low noise an important characteristic in a breast pump. Conversely, 55% of respondents are expected to belong to Latent Class 2, with a very high probability (95%) of considering low noise important. Further, Table 3 indicates members of both classes have a very high probability (90% and 99%, respectively) of considering comfortability an important breast pump characteristic.
Appendix B outlines parameter estimates for the remaining models. In order to better visualize item-response probabilities, characteristics were marked with a checkmark (ü) if there was a high probability of an Important response within a class (≥ 60%), a double dash (--) if there was a 50-59% probability, and cells were left blank if there was a low probability of an Important response (< 50%). Tables 4 through 7 use these visual indicators to graphically display the importance of breast pump characteristics to each latent class of mothers in the two-through five-class models.    In addition to the γ-parameters and ρ-parameters per model, model selection tools were considered to identify the final model (see Table 8). The AIC and BIC values are lowest at the two-class model, suggesting that this model is the best fit among these models; however, AIC and BIC bias towards smaller models based on their equations, thus exploring G 2 and BLRT is required. The drop in G 2 relative to degrees of freedom provides an improvement in fit for the two-class model. The four-class model was considered with a borderline significance of BLRT (p=0.08), but identification plots indicated convergence on a local log-likelihood maximum instead of the global, highest log-likelihood value (see Appendix B).
Finally, the only significant outcome of the BLRT occurred from the one-class to twoclass model (p=0.01), confirming the selection of the two-class model. Inspecting the parameter estimates from the two-class model suggests that the two classes are distinguishable and nontrivial.
One group of breast pumping mothers does not consider low weight, low noise, or discreet to be important characteristics of a breast pump (Latent Class 1), and the other group considers all seven breast pump characteristics important (Latent Class 2).
The two-class model was chosen as the final, baseline model, and the classes were titled "Form Follows Function," and "Wanting Everything," respectively.

LCA with Covariates
Mothers' working status, age, education level, and breast pump model were used as covariates on the two-class model. Age was bifurcated by birth year at the median (1983), with the Older group including those born in the year 1982 and before, and the Younger group including those in the year 1983 and later. Age was found to be a significant covariate (p=0.0106), while mothers' working status, education level, and breast pump model were not (p=0.6577, p=0.5615, and p=0.5950, respectively). Odds ratio plots (95% confidence interval, see Figure 1) show that the Younger group has higher odds of membership in the Form Follows Function class (noted in Figure 1 as "Class 2") relative to the Wanting Everything class. A significant covariate is indicated in Figure 1 by the fact that the confidence interval (rectangle) does not overlap with the y-axis value of 1; hence illustrating Age as a significant covariate.

LCA with Grouping Variables
The two-class baseline LCA model was explored to include observed groups as grouping variables. To test for measurement invariance between groups, the model was first fit with free ρ-parameter estimation, and then with restrictions that equate the ρ-parameters across groups. The model fits were compared and were not found to be significantly different (p=0.6845), which provides evidence that measurement invariance holds and indicates classes have the same meaning for each group. First, mothers' working status was included as a grouping variable with two levels: WOH and NWOH. Second, as mothers' age was found to be a significant covariate, mothers age in addition to working status was included as a grouping variable with four levels: WOH-Older, WOH-Younger, NWOH-Older, NWOH-Younger.
The optimal two-class model with mothers' working status as a grouping variable yielded the γ-estimates shown in Table 9. These γ-estimates indicate that there is almost an even probability that members of the WOH and NWOH groups belong to either of the two latent classes.
The baseline two-class model with mothers' age in addition to working status as a grouping variable yielded the γ-estimates shown in Table 10.

DISCUSSION
These results suggest that there are two distinct user groups of mothers in this sample who consider different breast pump characteristics important. The Form Follows Function group is solely focused on a breast pump's functionality, thus nonfunctional aspects of the pump (low weight, low noise, discreet) are considered unimportant. The Wanting Everything group considers each of the seven listed characteristics important in a breast pump, which contrasts with the summary statistics results indicating discreet was considered unimportant. Further, results indicate that membership in these groups is informed not by mothers' working status or age alone, but in fact by the two together. This provides evidence that mothers' needs vary for more complex reasons than simply whether or not they work from home. Similarly, the fact that the latent class analysis revealed further information regarding the importance of breast pump characteristics than the summary statistics suggests that quickly categorizing mothers based on their work status or another demographic characteristic is inaccurate for capturing mothers' breast pump needs.
The sample population reported similar breast pump experiences to the experiences described in the literature. Mothers reported nipple pain (48%), nipple damage (14%) and general discomfort and pain when using a breast pump (49%).
Mothers also reported feeling that breast pumping takes too long (79%) and that the pump is difficult to clean (43%) which directly correlates to the literature. This similarity between sample population experiences and experiences outlined in the literature further supports the validity of these two identified user groups. Identifying the different needs of mothers and sorting into accurate user groups, beyond simply demographic characteristics, will enable redesigns of breast pumps that address those user needs. Literature suggesting the benefit of individualizing breast pump recommendations to mother-infant dyads supports this conclusion (Meier, Patel, Hoban, & Engstrom, 2016).

Limitations
The survey did not ask mothers to rank breast pump characteristics from least to most important, which would have indicated what characteristic was most important to mothers, and did not allow mothers' free response. Also, the survey did not explicitly ask mothers whether this was their first infant, which could have been more of a predictor of breast pump experience than mothers' age.
Additionally, responses of importance were dichotomized in order to address scarcity at the 1 and 2 Likert levels. Expanding the surveyed population could resolve this scarcity and allow for the Likert levels to be analyzed individually, which could potentially alter the number and distribution of latent classes.
Further, this sample population was significantly skewed. While the sample had a slightly higher percentage of WOH mothers compared to the U.S. population (79% and 70%, respectively), the main reason for the skewness was education level (DeWolf, 2017). In this sample, 99.6% of respondents completed at least some college and higher, with 39% completing college and 52% completing graduate studies. This  Maternal andChild Nutrition, 12, 291-298. doi:10.1111/mcn.12137 Goodall, P. (1983). Design and gender. In The block reader in visual culture.
Appendix B: Detailed Results